| As one of the precision components inside rotating machinery,rolling bearing is widely used in large complex equipment such as wind power,tilting mechanism of converter and highspeed railway due to its advantages of high rotation accuracy and small starting torque.Because the bearing structure is complex and the operating condition is harsh,it is easy to appear such as pitting,cracking,rolling body spalling and other faults,leading to equipment failure.Therefore,the running condition monitoring and remaining applicable life(RUL)prediction of bearings are of positive significance for ensuring the safe and reliable operation of equipment and reducing the occurrence of major accidents.This paper takes rolling bearings as the research object,discusses vibration signal pretreatment and bearing RUL prediction methods,including the following three aspects.(1)According to the common threshold selection principle,the influence of decomposition scale factors on the noise reduction effect is not fully considered,and the soft threshold processing function has constant deviation,which restricts the noise reduction performance of the wavelet noise reduction method,an improved wavelet threshold noise reduction method is proposed.Based on the hard threshold function,an adjustable threshold function with parameters is proposed and the decomposition scale factor is taken into account in the process of noise reduction.Based on Hertz contact theory,a 2-degree-of-freedom dynamic model of outer ring fault of rolling bearings was established to obtain the simulation signals of faulty bearings,considering the random sliding of rolling body itself and time-varying displacement excitation caused by faults.The vibration signal of the faulty bearing in the outer ring is collected by the mechanical failure simulation test bench.Noise reduction experiments are carried out under simulated and measured bearing fault signals.The experimental results show that the proposed method can not only significantly improve signal quality,but also effectively reduce signal reconstruction errors.(2)In order to solve the problem that the accuracy of bearing RUL prediction is not high due to the distortion of degradation feature distribution under the condition of limited lifetime monitoring data of rolling bearings,a bearing RUL prediction method based on EWM and SVR is proposed.First,the time domain and frequency domain features of vibration signals are extracted,and the logarithmic transformation of the features is carried out.The index weights are determined by EWM to achieve feature selection.The SSA algorithm was used to optimize the SVR model.The low-dimensional features after dimensionality reduction by principal component analysis were taken as the input of the optimized SVR model,and the percentage of remaining service life was taken as the output to realize the prediction of bearing remaining life.The analysis results show that,compared with other methods with limited monitoring data,the proposed method not only has more stable prediction performance,but also reduces the absolute error by 19.51% and the mean square error by 17.73% on average.(3)As the traditional RUL prediction method requires a large amount of prior knowledge as support in the process of feature extraction and selection,and the feature extraction module and prediction module are separated from each other,the model parameters cannot be updated through error transfer between modules,which leads to the problem of low prediction accuracy in the case of sufficient data.An end-to-end bearing RUL prediction method combining gated CNN and convolutional Transformer encoder is proposed to add a convolutional layer inside Transformer,which further enhances the model’s ability to capture sequence information and realize accurate bearing RUL prediction.Through experiments and comparison with other RUL methods,it is verified that the proposed method can not only greatly reduce the prediction error,but also improve the generalization ability and robustness of the model. |
PDF Full Text Request